Predicting Outcome of NBA Games Using Artificial Neural

Predicting Outcome of NBA Games Using Artificial Neural Network Yung-Hsien Chu

Background National Basketball Association (NBA) is a popular men’s professional basketball league that includes 30 clubs. Many experts, gambling websites, and even fans themselves are trying to make prediction to NBA games.

Development of Prediciton Methods Originally, gamblers predict NBA games with their own experiences and instincts. In recent years, especially after the MLB Sabermetrics mania, experts and gamblers started to develop similar method and using some form of computer models to support their predictions.

Project Goal This project aims to predict outcome of NBA games based mostly on the statistics numbers, be it traditional statistics like point per game (PPG), or advanced statistics like True Shooting Percentage (TS%). Other possible features including injury status, team chemistry, and fatigue factor.

Data Collection All data extracted from basketball-reference website. The reason I choose this site is because this site provides not only traditional data, but also advanced data such as TS% and Turnover Percentage (TOV%). Another advantage of this website is it allows its data to be converted to CSV form for data processing.

Neuron Network Choice Back-propagation Supervised multilayer perceptron learning – win/lose outcome classification Non-linear Flexibility to process large amount of data.

Difficulties encountered so far There are many non-statistic factors that can have a great impact on the game outcome: injuries, rumors, and “team chemistry”. Team composition changes a lot between each year or even during basketball season due to trade or contract operation. Even though there advanced statistics, many are still questioning whether or not they can represent the ability of a player or team. Basketball is a sport that involving heavy interaction between players, so far there is no related data to represent it.

Project Expectation 58% accuracy should be expected, since it is the common percentage a normal gambler could achieve. 70% accuracy would be the ultimate goal, best analyzers and gamblers are known to have accuracy around 70%. (ESPN: 68. 3%, Bob Voulgaris: Around 70%)

Reference Miljkovic, D. , et al. "The use of data mining for basketball matches outcomes prediction. " Intelligent Systems and Informatics (SISY), 2010 8 th International Symposium on. IEEE, 2010. Yang, Jackie B. , and Ching-Heng Lu. "PREDICTING NBA CHAMPIONSHIP BY LEARNING FROM HISTORY DATA. " (2012). Zhang, Tao, Gongzhu Hu, and Qi Liao. "Analysis of offense tactics of basketball games using link prediction. " Computer and Information Science (ICIS), 2013 IEEE/ACIS 12 th International Conference on. IEEE, 2013.